1. Field of the Invention
The present invention relates to a wavelet processing apparatus and a related technique thereto.
2. Description of the Background Art
In JPEG2000 format, as shown in FIG. 57, an original image is decomposed into a plurality of rectangular segments in a matrix so as to execute a processing, each of which is called a tile 1. By executing the processing in the form of the plurality of tiles 1, the unit for one processing is reduced, so that the processing can be executed by means of hardware or software more easily. In particular, as for the hardware, this is effective in reducing the circuit scales of parallel process or LSI.
Basically, the JPEG2000 standard defines that a wavelet transform is executed at the time of image compression. As shown in FIGS. 58 to 60, in order to sequentially execute the wavelet transform, first as shown in FIG. 58, a predetermined region (such as entire frame or each tile 1) of the current image is considered as an object region 2a, and a plurality of lines 4a having a horizontal length which is equivalent to a horizontal size 3a of the object region 2a are prepared for a notice line 5a. Then in the object region 2a, while the notice line 5a is shifted in the vertical direction, each data (pixel) in this notice line 5a is subjected to filtering (high-pass and low-pass) in the vertical direction, followed by filtering (high-pass and low-pass) in the horizontal direction (scanning direction), for extracting high components and low components in each of the vertical direction and the horizontal direction to achieve the Mallat-type band decomposition. At the next stage (second stage), a region where the low components have been confirmed in both of the vertical direction and the horizontal direction is considered as an object region 2b, and as shown in FIG. 59, a plurality of lines 4b having a horizontal length which is equivalent to a horizontal size 3b of the object region 2b are prepared for a notice line 5b. Then while the notice line 5b is vertically shifted within the object region 2b, each data is subjected to filtering in both of the vertical direction and the horizontal direction (high-pass and low-pass) in the same manner as described above, to thereby achieve the Mallat-type band decomposition. Further, at the next stage, (third stage), a region where the low components have been confirmed in both of the vertical direction and the horizontal direction is considered as an object region 2c, and as shown in FIG. 60, while a notice line 5c which is selected from a plurality of lines 4c as shown in FIG. 60 is shifted in the vertical direction within the object region 2c, each data is subjected to filtering in both of the vertical direction and the horizontal direction (high-pass and low-pass) in the same manner as described above, to thereby achieve the Mallat-type band decomposition. Such filtering is performed on the data of a region where low components are confirmed in both of the vertical direction and the horizontal direction, and Mallat-type band decomposition is repeatedly executed.
In such a line-based wavelet method, however, it is necessary to process the respective notice lines 5a to 5c shown in FIG. 58 to FIG. 60 while storing them in a plurality of line memories as hardware. In such a process, if the line-based wavelet process is directly executed on the current image, the horizontal size 3a of the current image (FIG. 58) becomes too large when the image is such a large image that consists of as many as several millions of pixels (e. g., images taken by a digital camera), so that the memory capacity of line memory required for executing the wavelet transform becomes large. Therefore, conventionally, when such a large image that consists of as many as several millions of pixels is to be processed, as described above, the current image is decomposed into a plurality of tiles 1, and the line-based wavelet process is performed for each tile 1. By decomposing (tiling) into tiles 1 in the manner as described, it is possible to reduce the total memory capacity since only a plurality of line memories which are necessary and sufficient for the tile size which is a sub region are required for implementation.
However, there is a disadvantage that in the case of performing the wavelet transform using the plurality of tiles 1 as described above, the smaller the value of compression ratio (%) (i. e., the file size after compression is set smaller), a distortion occurs at the boundary of each tile 1 and a rectangular distortion becomes conspicuous.
By the way, considering the case that the object region 2a is defined in the manner that one frame of the current image is a single tile 1 for processing in order to avoid a distortion from occurring at the boundary of tile, this provides the same situation as the situation where tiling is not executed, so that the memory capacity of the line memory for executing the wavelet transform becomes extremely large as described above.